Flowly
Log inDownload
Use cases/Specialized

X Account Qualitative Analysis

Get an honest, qualitative read on your X account — voice patterns, what makes posts perform, audience signals — beyond the basic analytics dashboard.

Specializedmedium~30m setup
Tools
web_fetchmemory_search
Plugins
x-tools
Channels
telegramdesktop

X's built-in analytics tells you which post got the most likes. It doesn't tell you why. A qualitative analysis answers questions that move the needle: what topics consistently work, what voice tics get ignored, what's the gap between your posts and your audience's interests. Useful before big pushes (book launches, product campaigns, hiring) when getting it right matters.

What it does

  • Fetches your last N posts (100–500 typical)
  • Categorises by topic, format (single, thread, image, link), tone
  • Identifies patterns:
    • Topics that consistently outperform their average
    • Posts that underperformed despite similar topics — what differs?
    • Time-of-day / day-of-week effects
    • Followers-by-topic (who you attracted with each kind of post)
  • Output: a structured report you can act on

What you'll need

Setup

1. Run the analysis prompt

Send to Flowly
Analyse my last 200 X posts. For each post pull: - Text, timestamp, has_image, has_video, is_thread, is_reply - Likes, retweets, replies, impressions - Topic (you classify into 3-5 buckets you discover from the corpus) Then produce: ## Voice patterns - 3 phrases or moves I use that consistently work (with example posts) - 3 patterns that consistently underperform (with example posts) ## Topic performance Table per topic: post count, avg engagement, top post, worst post ## What outperforms its baseline Take posts where engagement is 2x my median for that topic. What's the common thread? List 3-5 hypotheses. ## What underperforms Posts < 0.5x median for their topic. What killed them? Hypotheses. ## Audience inference Who shares / replies the most? What does that tell me about who I'm actually reaching vs who I think I'm reaching? ## Action items 3 specific changes I should test in the next 20 posts. Save full analysis to memory tagged "x-analysis:<date>".

2. Re-run quarterly

Send to Flowly
Cron "x-analysis-quarterly" first day of each quarter: Run the same analysis on the last quarter's posts. Compare to the previous quarter's analysis (in memory). What shifted? Send Telegram message with delta.

3. Pre-campaign sanity check

Before you launch something — book, product, hiring round — run the analysis filtered to posts on that topic over the last 6 months. The agent surfaces what worked / didn't, you adjust the campaign messaging.

Example output highlights

Voice patterns

Works: opening with a counter-intuitive claim ("Most plugin systems fail because…") gets 2.3x median engagement.

Underperforms: hedged opens ("It might be argued that…"). Median engagement; reads as filler.

Topic performance

Plugin architecture posts (12 posts): avg 340 likes, top 1.2k AI evals posts (8 posts): avg 180 likes, top 320 Personal updates (15 posts): avg 95 likes

Insight: plugin content is your highest-leverage topic by a wide margin. AI evals are mid; personal posts are low. Lean into plugin content if reach is the goal.

Tips

  • Don't run weekly. Patterns need 3–6 months of data to be real. Quarterly is right; monthly is noisy.
  • Rate-limit reality. X API has aggressive rate limits. Pulling 500 posts may take an hour. Plan the analysis run for off-hours.
  • Watch for survivorship bias. Your wins are easier to remember than your flops. Force the agent to surface BOTH — a 3-week post that flopped is just as informative.
  • Don't optimise for everything. Pick one metric (impressions, follower growth, retweets) and tune the analysis around it. "Engage more" is too vague.
  • Voice changes drift. If the analysis says "this phrase underperforms" and you actually like saying it — keep saying it. Your voice matters more than 10% engagement.